It's a testament to the improvements in software engineering that we can even write large-scale systems at all in the first place. Structured programming and object-oriented programming [1] have proven to be revolutionary in the language front. Similarly, you can point to the development of the modern DAG-based VCS and continuous integration on the infrastructure front. Or modern IDEs with integrated debuggers and refactoring tools for development; automated regression suites; fuzzers and sanitizers--we've come a long way.

But I think a lot of people keep falling into the same trap when it comes to imagining what the next set of major improvements should be. Graphical programming is a popular example: the author pooh-poohs the terminal, for example. But text is the densest way we can precisely convey information; if we're trying to scale to the next system size, cutting information density by ½ (or more!) is not a good first step.

[1] I'm referring here specifically to the principle that state should be contained with objects that are protected by interfaces and passed around to people who need access to them, instead of everything just being functions that manipulate global variables. This is closer to what Alan Kay's original definition than the current interpretation.